International Journal of Computer Science and Software Engineering (IJCSSE), Volume 4, Issue 4, April 2015
ISSN (Online): 2409-4285
www.IJCSSE.org
Page: 95-101
Analytical Customer Relationship Management for Garage Services
Recommendation Using the Generalized Sequential Pattern Method
(A Study Case: PT. Armada International Motor)
Nina Setiyawati1, Wiranto H. Utomo2 and Danny Manongga3
1, 2, 3
Information System Department, Faculty of Information Technology, Satya Wacana Christian University, Salatiga,
Central Java 50711, Indonesia
1
[email protected], [email protected], [email protected]
ABSTRACT
Analytical CRM refers to CRM’s component needed by PT.
Armada International Motor ( PT. AIM ) Indonesia in achieving
customer knowledge that consists of customer behavior and
profile from PT. AIM’s garage database. Using data mining
exploration held on garage data in order to find possible pattern
as management’ precise decision making base. Generalized
Sequential Pattern (GSP) Algorithm refers to sequential data
mining algorithm which able in exploring data that able solving
any time and taxonomy constraints. GSP Algorithm implemented
in this study in order to find customer service pattern both
sequentially and simultaneously. One advantage of service type
patter founded that taken simultaneously is as service type
recommendations base that can be run simultaneously at one
service period. Furthermore, it may be used as promotion
planning base, e.g. product bundling.
Keywords: Analytical CRM, Data mining, GSP, Garage.
1. INTRODUCTION
PT. Armada International Motor (PT. AIM) is one of the
automotive distributor companies in Indonesia which has
selling division and garage division to provide services for
customers’ vehicles.
According to the data, the growth of Indonesia's car
market has always experienced a rising trend from 2010
[1]. This fact makes the competition in the market more
competitive, so that every automotive distributor company
is competing for the market. According to Ross [2], a key
factor to gain market share is to build customers’ loyalty
and develop a sustainable competitive advantage.
Keeping the customers and increasing their loyalty is two
very important financial activities within a company.
Based on the fact that attracting new customers is much
more costly [3][4]; therefore, studying the factors that
support the growth of the loyalty of the customers
becomes a crucial management issue. This has encouraged
PT. AIM to improve the strength of the relationship with
customers for the company's development process and
increasing the profits by implementing CRM (Customer
Relationship Management). However, only the new
operational CRM system is applied in PT.AIM. This has
led to the company’s superficial knowledge of the
consumers’ profile and behavior, resulting at the company
cannot predict future actions. Therefore, it is necessary to
have an analytical CRM system that focuses on the
processing and interpretation of the data collection that has
been stored in order to create meaningful and beneficial
interactions with customers [5]. Analytical CRM is the
most critical component in CRM [6] because it determines
the ability of the CRM system to acquire knowledge of
customers [7] and allows to effectively manage
relationships with customers [5].
From the database of PT. AIM’s garage we can obtain
database, customer data, service transactions, and the types
of services performed. However, it is still unknown how
the behaviors of customers are in choosing the types of
service in each scheduled service. This will result in the
random selection types of service. Types of service and the
schedule of service is an important factor in taking care of
the vehicle. There are several types of services that are
related to each other, and the schedule of service is an
important factor in determining the type of service that is
performed in a sequential manner. Therefore, it is
necessary to apply the data mining techniques to explore
patterns of the service done by the customer.
The implementation of data mining techniques to identify
opportunities and optimize customers’ interactions [8] is
an analytical CRM step which is a process associated with
the use of data effectively, efficiently, and strategically, so
as to enable the right decisions for the management [9].
Data mining can also search for the compatibility between
the product and the customers and provide better targeting
International Journal of Computer Science and Software Engineering (IJCSSE), Volume 4, Issue 4, April 2015
N. Setiyawati et. al
of promotional products. In other words, it can be said that
data mining helps CRM in their responsibility to obtain,
develop and maintain customer’s loyalty [10].
One method of data mining is generalized sequential
pattern (GSP), which is a method that can look for
relationships between different items in a dataset where
different data are on a transaction that results in a different
grouping. The data of what kind of service transactions
that already exist in the database will be excavated and
grouped to find a distribution pattern of the entire dataset;
and there will be a result of a service relationship pattern
that is performed simultaneously and also a type of service
pattern that is performed sequentially.
Based on the description above, this study will apply the
algorithm of generalized sequential pattern to the data of
PT. AIM’s garage in order to determine the pattern of
relationship of the types of services that are performed
simultaneously or sequentially, which is used as the basis
for service recommendations and basic managerial
decision making in improving relationships with
customers.
2. RELATED WORK, RESEARCH
PURPOSE AND CONTRIBUTION
2.1 Related Work
A study entitled "Implementation of Data Mining
Techniques in The Concept of Customer Relationship
Management (CRM)" explains that there are three major
types of CRM, one of which is an analytical CRM that
enables an accurate decision-making action for the
management because it involves the process of analysis,
modeling, and evaluation of the data stored in the database
to generate a mutually beneficial relationship between the
company and its customers. In this study, the data mining
applied in CRM concept uses the association rule method
to obtain the association between one product and another
product as the basis of marketing strategies for cross
selling in retail companies [11].
The study "Data Mining with Fuzzy Method for Customer
Relationship Management (CRM) in Retail Company" [12]
discusses about the data mining process from the sales data
of UD. Fenny, a retailer of baking materials and equipment
located in Denpasar, Bali, looks for potential customers to
conduct consumer segmentation which is the process to
determine consumers’ behavior and implement appropriate
marketing strategies to be profitable for the company. The
data mining process begins with the process of clustering
using Fuzzy C-Means (FCM) algorithm and Fuzzy
Subtractive (FS) Clustering. The clustering results of each
of the methods are used for segmentation using Fuzzy
RFM model to find the consumer class.
96
2.2 Research Purpose
In this study, extracting the data of the service transactions
in the garage is conducted to obtain customer transaction
patterns either simultaneously or sequentially using the
GSP algorithm as an CRM analytical step which is used as
the basis for giving recommendations of service and the
basis for managerial decision making in improving
relationships with customers.
2.3 Research Contribution
For the company, the use of analytical CRM can give
knowledge in the form of customer transaction patterns
that can be used by the management as a basis for
decision-making actions in maintaining customers’ loyalty,
such as promotions of product building or product affinity,
and the company’s CRM application development which
is based on the knowledge found. This will give a positive
effect not only for the garage division, but also for the
sales division.
In the future, this study is expected to contribute to the
learning of analytical CRM and development of CRM
applications.
3. GENERALIZED SEQUENTIAL
PATTERN ALGORITHM
Generalized Sequential Pattern (GSP) is a sequential data
mining algorithm that can overcome the limitations of time
and taxonomy introduced by R. Srikant and R.Agrawal on
EDBT'96. GSP algorithms perform several processes of
the data to dig deeper into the data in order to find
knowledge, as illustrated in the process of extraction on
Knowledge Discovery in Databases in Figure 1.
Fig. 1. Knowledge Discovery in Database Process [13]
Based on the process of knowledge discovery in
database in Figure 1, before the process of extracting the
data, data preparation is performed. The purpose of the
97
International Journal of Computer Science and Software Engineering (IJCSSE), Volume 4, Issue 4, April 2015
N. Setiyawati et. al
data preparation process is to prepare the data into an
effective and efficient form to be extracted. The data
preparation process in this study is divided into three main
stages, which are:
1. Data selection
Data selection is a process of selecting the data from the
garage of PT. AIM in the year of 2009 to 2014 which
consists of 5070 customers, 10610 transactions and 59400
detailed transactions. The relevant data are selected to
obtain the target data, such as transaction data, the data of
spare parts, the data of the type of service, and the dates of
service.
The following GSP algorithm process in this study are:
A. Finding Individual Item with minSupport (1Sequence)
Before finding individual item, minimum support must be
determined first, which in this study is 0.5. Will be
founded the candidate from data sequence or database that
fulfilled the minimum support. After candidate 1sequence founded, system will calculate the number of
candidate have been obtained from sequence data. This
process named as counting candidate thus resulted
frequent 1–sequence as shown at Table 2 (Process shown
with item id number in database).
2. Preprocessing
Preprocessing is the process of consolidating the data
targets with a special structure to make it more efficient.
The table attributes that are not used will be omitted to
make them more efficient and form a new relationship of
tables that have been established. The data of spare parts
and the data of the types of services are included in one
table.
3. Transformation
In this process, the results of the preprocessing will be
transformed into an appropriate form of database to be
extracted. The results of this transformation process are
item table, transaction table, transaction_detail table, and
time table.
4. Result and Discussion
After the data passed the preparation process, then it will
proceed to the data mining process which is the process of
extracting data to find patterns in the data set that has been
established. In this study, the data collection was done by
using the GSP algorithm. Before carrying out the GSP
algorithm, predetermination of the time constraint was
conducted to limit the time gap between the series of
transactions that contained sequential elements on the data
sequence [14]. Table 1 shows the time constraint in this
study.
Table 1: Time Constraint
Time Gap
1
2
3
…
Start Time
2009-01-01
2009-04-01
2009-07-01
…
End Time
2009-03-31
2009-06-30
2009-08-31
…
Table 2: Frequent 1-Sequence
Frequent
1-sequence
51383
46472
47072
41006
40956
50259
40957
48633
…
Number of
Item
7109
3733
7412
7410
7625
6387
7622
5973
…
B. Using The Individual Item in Finding 2-Sequence
From individual item at frequent 1-sequence, merging and
pruning processes held which are the part of candidate
generation process. Candidate generation results in this
phase shown at Table 3.
Table 3: Candidate Generation 2-Sequence
Frequent 1sequences
51383
46472
47072
41006
40956
50259
40957
48633
…
Candidate 2-sequence
after join
after pruning
(6472, 8633)
(6472, 8633)
(0956, 6472)
(0956, 6472)
(0957) (1006)
(0957) (1006)
(0259) (1006)
(0956) (8633)
(0956) (8633)
(1006) (0259)
(1006) (0259)
(0957) (0259)
(0957) (0259)
(1006) (8633)
(1006) (8633)
(1006, 8633)
(1006, 8633)
…
…
After candidate generation process, then will be held
counting candidate process. Counting candidate results in
this phase are shown at Table 4.
98
International Journal of Computer Science and Software Engineering (IJCSSE), Volume 4, Issue 4, April 2015
N. Setiyawati et. al
Table 4: Frequent 2-Sequence
Frequent 2-sequence
(46472, 48633)
(50956, 46472)
(40957) (41006)
(40956) (48633)
(41006) (50259)
(40957) (50259)
(41006) (48633)
(41006, 48633)
…
Number of Item
4553
4816
4940
4559
4681
4559
…
Table 5: Candidate Generation 3-Sequence
(40957) (41006)
(41006) (50259)
(40957) (50259)
(41006) (48633)
(41006, 48633)
…
…
(41006)
(40957)
(48633)
(41006)
…
Table 6: Frequent 3-Sequence
After frequent 2-sequence founded, then candidate
generation 3-sequence is held. Candidate 3-sequence
results shown at Table 5.
(40956, 46472)
(40957)
(48633)
Candidate 3-sequence will be calculated in resulting
frequent 3-sequence. Frequent 3-sequence shown at Table
6.
C. Using K-Sequence to Find (K+1)-Sequence
Frequent 2sequences
(46472,48633)
(40956) (8633)
Candidate 3-sequence
after join
after pruning
(40957)
(41006, (40957)
(41006,
48633)
48633)
(40956,
46472, (40956,
46472,
48633)
48633)
(40957)
(41006) (40957)
(41006)
(50259)
(50259)
Frequent 3-sequence
(40957) (41006, 48633)
(40956, 46472, 48633)
(40957) (41006) (50259)
(40957) (41006) (48633)
…
Number of Item
4811
4810
4553
4558
…
This process will retain. Itemsets at frequent 3-sequence
used in resulting candidate 4-sequence.
D. Finding The Last Frequent Sequence
In this study, process will be ended up after frequent 8sequence found. Frequent 8-sequence results on
application are shown at Figure 2.
Fig. 2. Sequential Pattern of Data Mining Application Result
Figure 2 is sequential pattern of data mining result or in
this study refers to itemset at frequent 8-sequence that
including some sequential patterns. This are following half
of sequential patterns found:
•
{(Ganti oli lengkap dan of) (Ganti oli mesin)
(Engine tune up - V/C/A) (ATF dextron II) (Elf
tranself type B) (Fuel filter, Isuzu genuine oil)
(Alumunium washer)}
International Journal of Computer Science and Software Engineering (IJCSSE), Volume 4, Issue 4, April 2015
N. Setiyawati et. al
•
{(Brake 4 wheel – V/C/A) (Ganti oli lengkap dan of)
(Ganti oli mesin) (Engine tune up-V/C/A) (Elf
tranself type B, Air accu, Kuras minyak kopling)
(Fuel filter, Isuzu genuine oil) (Alumunium washer
14-22)}
•
{(Brake 4 wheel – V/C/A) (Ganti oli lengkap dan of)
(Ganti oli mesin) (Engine Tune Up-V/C/A,
Bleeding oli rem) (Elf tranself type B) (Air accu,
Alumunium washer)}
•
From several sequential patterns found, it can be
seen that :
•
Except Engine tune up - V/C/A , Ganti Oli mesin is
type of service that always included in pattern.
•
After doing Ganti oli mesin, Engine tune up V/C/A will be taken on the next period.
•
From 8 available pattern, Elf tranself type B spare
part purchasing will be held, after Engine tune up V/C/A done.
99
According to CRM dimension consists of customer
identification, customer attraction, customer retention and
customer development [8], sequential pattern of service
type found may be used as the basic in performing one of
CRM task in customer development dimension, which is
sequential market based analysis or next sequential
purchase that belong to analysis in forecasting what
product or service that will be bought by customer in
further [15]. In this study, the main aim of sequential
pattern of service type that found is as the basis of
providing type of service recommendation that must be
taken in the next service. While the other advantage is as
promotion base by making type of service package
sequentially in a year supported with membership system
thus customers will get discounts or cheaper price instead
choose type of service by themselves at each visit.
In addition to type of service pattern that taken
sequentially, by GSP algorithm also found type of service
pattern that taken simultaneously. Figure 3 shows type of
service that taken simultaneously
Fig. 3. Association Pattern of Data Mining Application Result
Figure 3 is association pattern or type of service
pattern that often taken simultaneously. Association
patterns that fulfilled the minimum support were found,
some of which are:
• {(Engine tune up – V/C/A, Elf Tranself Type B,
Fuel filter, Air cleaner, Isuzu genuine oil, Air accu)} with
2756 number of transactions.
• {(Ganti oli lengkap dan of, Engine tune up –
V/C/A, Elf Tranself Type B, Fuel filter, Air cleaner, Isuzu
genuine oil )} with 2604 number of transactions.
Of several association patterns found, can be seen
that :
• Type of service Ganti oli mesin dan of is often
taken simultaneously with Engine tune up - V/C/A.
• Type of service Engine tune up - V/C/A is often
taken simultaneously with combination of Elf tranself type
B Fuel filter, Air cleaner, Isuzu genuine oil, and Air accu.
• Fuel filter, Air cleaner, Isuzu genuine oil, and Air
accu are often purchased simultaneously.
• Spare part combination that sold simultaneously
has greater sale number than type of service combination
that taken simultaneously.
Association pattern of type of service that found is the
result of CRM analytical that may be used in performing
one of the CRM’s task in customer development
dimension, namely market based analysis [8], especially
product bundling. The main aim of this study of
association of service type that has been found is as the
International Journal of Computer Science and Software Engineering (IJCSSE), Volume 4, Issue 4, April 2015
N. Setiyawati et. al
basis of providing type of service recommendations that
possible to be taken simultaneously by customer.
Furthermore, another advantage is as the basic in (b)
providing information to management as the basic of
promotion making of type of service package, (c)
providing information to management as the basic to
create promotion of type of service package together with
spare part, (d) providing information to management as the
basic to create promotion of spare part purchasing package.
5. Conclusions
Based on the results of the study, it is concluded that the
Generalized Sequential Pattern algorythm was able to
excavate the data and determine the customers’ behavior
patterns in selecting the types of services conducted
simultaneously and even sequentially. There were several
sequential patterns of the datamining results that produced
some knowledges, one of which was after doing the Ganti
oli mesin, the Engine tune up - V/C/A would be the
service type in the following periode. The main purpose of
finding each sequential pattern was to be used as the basis
of providing the service type recommendation that had to
be done in the next service periode. Moreover, it can also
be used as the basis of promotion, i.e. to make a service
package for consecutively one-year periode using a
membership system, so that customers could get discounts
or lower prices compared to when they choose a service
type by themselves in every visit.
Data mining with GSP algorythm also generated
association patterns, one of which was the oil change
service type that was frequently done along with Engine
tune up – V/C/A. In addition, the main purpose of finding
each association pattern was as the basis for providing
recommendations on service types that could be
simultaneously given to customers. Besides that, it could
also provide information to the management as the basis of
making promotions of the service type packages, the
service package with spareparts, and the sparepart buying
package.
For further studies, this study can hopefully be used as the
source of application development, both web-based and
mobile applications, so that customers can get the access
in finding the recommendations of services that have to be
done and the reminders of service schedules.
REFERENCES
[1]Anonym, “Insentif Fiskal Untuk Meningkatkan Daya Saing
Industri Kendaraan Bermotor Roda Empat di Indonesia”,
Kementrian
Keuangan
Republik
Indonesia,
www.kemenkeu.go.id, 2013.
100
[2]D. F. Ross, “E-CRM from a Supply Chain Management
Perspective”, Journal of Information Systems Management,
Vol. 22, Issue 1, 2005, pp 37-44.
[3]A. Ghavami, and A. Olyaei, "The Impact of CRM on
Customer Retention", Masters thesis, Department of Business
Administration and Social Sciense, Lulea University of
Technology, Lulea, Sweden, 2012.
[4]J. Ranjan, "Role of Analytical CRM in CRM Systems:
Importance and Benefits", Management & Change, Vol. 13,
No. 1, 2009, pp. 69-88.
[5]V.L.M. Oliveira, "Management in Retailing Supported by
Data Mining Techniques", Doctoral thesis, Department of
Industrial Engineering and Management, University of Porto,
Porto, Portugal, 2012.
[6]J. Reynolds, "A Practical Guide to CRM: Building More
Profitable Customer Relationships", New York: CMP Books,
2002.
[7]M. Xu, J. Walton, “Gaining Customer Knowledge Through
Analytical CRM", Industrial Management & Data Systems,
Vol. 105, No. 7, 2005, pp. 955-971.
[8]R. M. Nasr, S. Nagaty, M. B. Senousy, "Survey and
Challenges of Mining Customer Data Sets to Enhance
Customer Relationship Management", International Journal of
Computer Applications (0975 – 8887), Vol. 107, No. 13,
2014, pp. 34-40.
[9]Y. Xie, X. Li, W. Ying, "A Process Driven Architecture of
Analytical CRM Systems with Implementation in Bank
Industry", International Journal of Intelligent Information
Technology Application, Vol. 1, No. 1, 2008, pp. 48-52.
[10]K. Tsiptsis, A. Chorianopoulos, "Data Mining Techniques in
CRM: Inside Customer Segmentation", Chicester: A John
Wiley and Sons, Ltd., 2009.
[11]
B.A. Tama, "Implementasi Teknik Data Mining di
Dalam Konsep Customer Relationship Management
(CRM)",Konferensi Nasional Sistem dan Informatika 2009,
pp. 56-61.
[12]Y. Nugraheni, "Data Mining Dengan Metode Fuzzy Untuk
Customer Relationship Management (CRM) Pada Perusahaan
Retail", Masters thesis, Department of Graduate Program,
University of Udayana, Denpasar, Indonesia, 2011.
[13]F. Usama, P. Gregory,P. Smith, "From Data Mining to
Knowledge Discovery in Database", AI Magazine, 1996.
[14]R. Srikant, R. Agrawal, "Mining Sequential Patterns:
Generalizations and Performance Improvements", San Jose:
IBM Almaden Research Center, 1996.
[15]J. Dyche,The CRM Handbook: A Business Guide to
Customer Relationship Management, USA : Addison-Wesley,
2004.
AUTHOR PROFILES:
Nina Setiyawati is a senior college in Information System
Department, Information Technology Faculty, Satya Wacana
Christian University, Salatiga, Indonesia. Despite studying in
order to achieve M.Cs degree tittle, she is also actively working
on Netniho Software House and as a lecture assistance in
Information Technology Faculty. Research specialization is in
field of business intelligence and software engineering.
International Journal of Computer Science and Software Engineering (IJCSSE), Volume 4, Issue 4, April 2015
N. Setiyawati et. al
Wiranto H. Utomo is a lecture and professor in Information
Technology Faculty, Satya Wacana Christian University. He got
master degree in Computer Engineering at Gajah Mada
University in 2002 and his PhD at Gajah Mada University in
2011. He has bear dozens of studies that published on IJCSI,
IJWA, MASAUM, Journals, IJCA, and in International
Conference including IWAS of the ACM. Research
specialization of Prof. Dr. Ir. Wiranto Herry Utomo, M.Kom. is
in the field of SOA, java EE, web service, business intelligence
and software engineering.
Danny Manongga is a lecture, professor and head of
Information System Department, Information Technology
Faculty, Satya Wacana Christian University. He got master
degree in Information Technology at Queen Mary CollegeUniversity of London in 1989, and PhD in Management
Information System and Operation Research at University of
East Anglia in 1996. Research specialization of Prof. Ir. Danny
Manongga, MS.c., Ph.D. is in the field of business intelligence,
artificial intelligent, knowledge management, and social network
analysis (SNA).
101